Catálogo de publicaciones - libros

When performing online speaker clustering, it is common to make clustering decision as soon as an audio segment is received. When the wrong decision is made, the error can propagate the posterior clustering. This paper describes a decision-tree-based online speaker clustering algorithm. Unlike typical online clustering approaches, the proposed method constructs a decision tree when an audio segment is received. A pruning strategy for candidate-elimination is also applied. Experiments indicate that the algorithm achieves good performance on both precision and speed. Finally, we discuss the relation between the performance and the width of the decision tree beam.

Recently, hidden Markov models (HMMs) have been found to be very effective in classifying heart sound signals. For the classification based on the HMM, the continuous cyclic heart sound signal needs to be manually segmented to obtain isolated cycles of the signal. However, the manual segmentation will be practically inadequate in real environments. Although, there have been some research efforts for the automatic segmentation, the segmentation errors seem to be inevitable and will result in performance degradation in the classification. To solve the problem of the segmentation, we propose to use the ergodic HMM for the classification of the continuous heart sound signal. In the classification experiments, the proposed method performed successfully with an accuracy of about 99(%) requiring no segmentation information.

In this paper a novel symmetric protocol to cipher digital images is introduced. The protocol proposed in this work is based on the paradigm stated by J. Fridrich at 1998. Consequently, there are two iterative stages in the algorithm: The confusion stage permutes the pixels in the image using the Cat map, whereas in the diffusion stage, the pixel values (the color of each pixel) are modified sequentially such that a small change in the color of only one pixel is spread out to many pixels. This second phase is carryied out by means of a reversible cellular automaton. The proposed protocol is shown to be secure against the more important cryptanalytic attacks.

Coarseness is a very important textural concept that has been widely analyzed in computer vision for years. However, a model which allows to represent different perception degrees of this textural concept in the same way that humans perceive texture is needed. In this paper we propose a model that associates computational measures to human perception by learning an appropriate function. To do it, different measures representative of coarseness are chosen and subjects assessments are collected and aggregated. Finally, a function that relates these data is fitted.

This paper proposes a system to estimate the 3D position and velocity of vehicles, from images acquired with a monocular camera. Given image regions where vehicles are detected, Gaussian distributions are estimated detailing the most probable 3D road regions where vehicles lay. This is done by combining an assumed image formation model with the Unscented Transform mechanism. These distributions are then fed into a Multiple Hypothesis Tracking algorithm, which constructs trajectories coherent with an assumed model of dynamics. This algorithm not only characterizes the dynamics of detected vehicles, but also discards false detections, as they do not find spatio-temporal support. The proposals is tested in synthetic sequences, evaluating how noisy observations and miss-detections affect the accuracy of recovered trajectories.

This paper outlines a neural network model based on the Fuzzy Cognitive Maps (FCM) framework for solving the automatic image change detection problem. Each pixel in the reference image is assumed to be a node in the network. Each node has associated a fuzzy value, which determines the magnitude of the change. Each fuzzy value is updated by a trade-off between the influences received from the fuzzy values from other neurons and its own fuzzy value. Classical approaches in the literature have been designed assuming that the mutual influences between two nodes are symmetric. The main finding of this paper is the assumption that mutual influences could not be symmetric. This non symmetric relationship can be embedded by the FCM paradigm. The performance of the proposed method is illustrated by comparative analysis against some recent image change detection methods.

Context-Free Grammars play an important role in the pattern recognition research community. Word graphs provide a compact representation of the ambiguous alternatives generated during many pattern recognition, machine translation and other NLP tasks. This paper generalizes the framework for string parsing based on semirings and hypergraphs to the case of lattice parsing. This framework is the basis for the implementation of a parsing interface in a dataflow software architecture where modules send and receive word graphs in a serialized form using a protocol which allows the easy generation, filtering and parsing of word graphs. An implementation of the CYK algorithm is presented as an example. Experimental results are reported to demonstrate the proposed method.

This paper presents a monocular perception system tested on wheeled mobile robots. Its significant contribution is the use of a single image to obtain depth information (one bit) when robots detect obstacles. The constraints refer to the environment. Flat and homogeneous floor radiance is assumed. Results emerge from using a set of multi-resolution focus measurement thresholds to avoid obstacle collision. The algorithm’s simplicity and the robustness achieved can be considered the key points of this work. On-robot experimental results are reported and a broad range of indoor applications is possible. However, false obstacle detection occurs when the constraints fail. Thus, proposals to overcome it are explained.